[argyllcms] Re: negative primaries

  • From: Ben Goren <ben@xxxxxxxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Sun, 6 Jun 2010 17:59:40 -0700

On 2010 Jun 6, at 1:51 PM, Elle Stone wrote:

> But really even ProPhoto isn't big enough for all images.

It's even worse than that, I'm afraid.

The spectrum locus lies at least partially outside both Lab and XYZ color 
spaces. That is, there are huge swaths of spectrally-pure colors that no color 
space in use today can encode without clipping.

The good news is that these colors are rarely encountered outside of lasers or 
tunable diffraction gratings. (And prisms.) The better news is that few input 
devices can capture these colors -- and no output device can even come vaguely 
close to reproducing them.

Some time back this whole point was driven home to me in a pretty emphatic 
manner when I tried to create a digital rainbow. I created a .ti3 file that was 
all zeros except for a single frequency at 100, and used fakeread to convert 
that to Lab colors. Much beyond the far red and violet portions of the 
spectrum, Argyll began reporting Lab values well beyond 100 L* and 127 a* and 
b*. So, after some experimentation, I kept the single peak at 100 but added 
values of 50 on either side. It brought the Lab values closer to within range, 
but still waaaay out. It eventually took a very wide-band bell curve to get 
something that would fit in Lab space -- and that, of course, was still waaaaay 
outside any other color space.

Attached is the end result in sRGB. (I think I just used Photoshop to convert 
from Lab to sRGB, so it's decidedly not the best possible representation, but 
not all *that* bad.)

Even aside from the banding from the very low resolution, it's pretty obviously 
not what one sees in a sparkling crystal. Yet it's probably about as close as 
you'll ever see on anything even vaguely resembling an LCD display, picture 
tube, or any similar device. A wide-gamut monitor would obviously do better, 
but even those are so far from being capable of creating spectrally-pure colors 
it's not even funny. And printers? Don't make me laugh.

When the pixels in our displays are made not of three primaries (or even twice 
that number), but are made from tunable diffraction gratings (or something 
along those lines), then it'll be time to worry about the limits of modern 
color spaces. Until then, it's just an exercise in futility.

For printing, barring some truly radically revolutionary advances in chemistry 
and physics, it'll never even pretend to be a problem.




Other related posts: